An evidence pack, illustrated
This is not a client. There is no engagement behind this page.
It is an invented example of what an AI Value & Governance Diagnostic produces, written so you can see the shape of the deliverable before you decide whether to talk to us. Every number describing this company was made up for the purpose. None of it is a benchmark, and none of it appears anywhere else on this site.
What is real is the method: what gets inspected, what turns out not to be inspectable, and how a finding becomes a decision somebody can defend in a board meeting.
An invented company. A real method.
ILLUSTRATION
What this fictional organization asked
Roughly three thousand people, four business units, AI in production for about two years. A board paper is due in eleven weeks asking for the next funding round.
The question, as the CFO's office put it: what has the current spend returned, and which parts of it deserve more?
Nobody inside disagreed that the question was fair. Nobody could answer it.
Five dimensions. Two came back with evidence
CAPACITY — assessed
Spend across fourteen line items in three cost centres, reconciled to invoices. Two of the fourteen were duplicate contracts for the same capability, bought nine months apart by different units.
ACCESS — assessed
Who can reach which models and which data, from the configuration as deployed rather than as documented. Four groups had access nobody could explain; two of them were leavers' groups that had never been closed.
PROVENANCE — could not be assessed
Nothing records where a model output came from, which version produced it, or what it was given. Not because anyone decided against it — no operational system does this unless somebody paid for it at build time.
COMPLIANCE — could not be assessed
The obligations are documented. The controls are happening. Nothing connects one to the other, so "are we compliant" can be asserted and cannot be demonstrated.
ATTRIBUTION — could not be assessed
Spend sits with finance, outcomes sit with four product teams, on different calendars against different units of work. Reconstructing the link after the fact was not expensive. It was impossible.
That is the finding
Three of five is not a failed review. It is the answer to the question the board is about to ask, arriving eleven weeks before the board asks it.
This organization cannot demonstrate what its AI spend returned. Not because the spend was wasted — some of it clearly was not — but because nobody built the evidence layer, and you cannot reconstruct one backwards.
Everything below follows from that.
Three of the five had nothing to inspect. That is the finding.
- CAPACITY — assessed
- ACCESS — assessed
- PROVENANCE — could not be assessed
- COMPLIANCE — could not be assessed
- ATTRIBUTION — could not be assessed
Eleven investments. Every one gets a verb
STOP — 2
The duplicate contract, and a pilot that has been ninety-percent complete for five quarters with no named owner.
SCALE — 1
One workflow in the service unit that is measurably faster and has a person accountable for the number. It is underfunded because nobody knew it was working.
FIX — 3
Three deployments where the value is real and the execution is not: no gate before output reaches a customer, no rollback, no second measurement after launch.
GOVERN — 4
Four are running with nobody owning the risk. Two touch customer data through a route the security review never saw.
FUND — 1
The evidence layer itself. It is the only item on the list that makes the other ten answerable next year.
The verbs are the deliverable
Not a maturity score, not a percentage, not a heat map. Eleven decisions a board can take, each with the evidence that supports it attached, and each with a named person who owns the next step.
An assessment that ends in a score ends in a conversation about the score.
Eleven investments. Eleven decisions. No score.
Illustrative — not a measurement
- STOP 2
- SCALE 1
- FIX 3
- GOVERN 4
- FUND 1
What the first ninety days say
Days 1 to 30 — close what is already decided
The duplicate contract. The four unexplained access groups. The stalled pilot. None of these needs a strategy; they need somebody to do them, and they were all closed inside a month.
Days 31 to 60 — the smallest evidence layer that answers the question
Not instrumenting everything. Picking the three decisions the board will actually ask about, and building the trail for those. This is the item marked FUND, and it is deliberately small.
Days 61 to 90 — the second measurement
The one workflow marked SCALE, measured again on the same definition as the baseline. Whatever that number is, it is the first defensible ROI figure the organization has ever had.
The board paper is due in week eleven. It goes in with one measured result, ten decisions with evidence behind them, and an honest statement about what still cannot be measured and what it would cost to fix.
What you will not find in here
No maturity score. No benchmark against a peer group we have not measured. No vendor recommendation — we hold no stake in any of the tools involved and we do not resell them.
No confident number for anything we could not inspect. Where the evidence does not exist, this pack says so and prices building it. That is the least satisfying page in any assessment and the only one that holds up.
We do not issue audit opinions, assurance opinions or certifications. A pack that implied otherwise would be worth less than nothing to you.
Every number describing this company is invented. None of it is research, vendor data, or anybody's measurement. → How we use numbers
Your organization is not this one. The five questions are the same.
Start a conversation about the decisions in front of you.